Spectral analysis with an exponential window iterative method: The performance of the signal frequency evaluation

The analysis of the transients of the voice signal by signal sampling with filter parameters readjusted at short intervals is discussed. An iterative method is used to compute the correlation coefficients after each sampling interval, simulating the computing of the autocorrelation in a digital filter with the voice signal as input, which allows a linear predictive coding analysis with good resolution and small computing times to be performed. A procedure to verify the rapidity and precision of the frequency estimation, using ramp modulated variable frequency test signals was adopted. The convergence of the coefficients of the iterative correlation is confirmed. The capacity of the algorithm to follow fast spectral variations is shown.